Design Principles

Go-Live's architecture is guided by three core principles that shape every design decision. These principles emerge from the WebRTC SFU design space and prioritize operational simplicity over feature completeness.

1. Simplicity Over Completeness

Single-publisher rooms eliminate the complexity of stream selection algorithms (SVC adaptation, simulcast layer selection) that dominate large-scale SFU implementations [1]. By constraining each room to one publisher, Go-Live avoids:

• Simulcast layer selection: No need for bandwidth estimation algorithms or temporal/spatial layer switching
• Stream prioritization: No arbitration logic when multiple publishers compete
• Codec negotiation complexity: Single publisher means single codec pipeline per room

This constraint is a deliberate trade-off. Go-Live targets use cases like live events, webinars, and surveillance where the one-to-many model is natural. For multi-party conferencing, projects like Janus or mediasoup provide more appropriate abstractions.

2. Zero-Transcoding Forwarding

By operating at the RTP packet level, Go-Live avoids the computational cost of decode/encode cycles. The SFU reads RTP packets from the publisher's track and writes them directly to subscriber tracks without any media processing [2].

What this enables:

• Predictable CPU usage (scales with subscriber count, not media complexity)
• No codec licensing concerns (H.264/VP8/VP9 pass-through)
• Sub-frame latency (no encode pipeline delay)

What this sacrifices:

• No server-side mixing or compositing
• No bitrate adaptation to subscriber conditions
• No server-side recording in formats requiring re-encoding (e.g., MP4)

The recording system works around this by capturing raw RTP payloads into container formats that support them: VP8/VP9 into IVF, Opus into OGG.

3. Observable by Default

Every significant operation emits metrics and traces, enabling production debugging without code changes. The observability stack consists of:

• Prometheus metrics: Room count, subscriber count, RTP bytes/packets per track, rate limit rejections
• OpenTelemetry traces: Request-level tracing through the HTTP handler, SFU manager, and room operations
• Health endpoints: /healthz for liveness, /metrics for Prometheus scraping

This principle ensures that when a production issue occurs, the first question is "what do the dashboards show?" rather than "can we reproduce this locally?"

Concurrency Model

Go-Live's concurrency design follows Go idioms:

Concern	Mechanism
Room state	sync.RWMutex per room
Track fanout	Goroutine per subscriber, exits on disconnect
Graceful shutdown	Context cancellation propagates to all goroutines
Rate limiting	Token bucket per IP, goroutine-safe

Each subscriber gets an independent readLoop goroutine that reads from the publisher's track and writes to the subscriber's local track. This goroutine exits when:

• The subscriber disconnects (ICE state change)
• The publisher disconnects
• The room is closed

See ADR-0001: Core Architecture for the full architectural decision record.

References

1. RFC 7667 - "RTP Topologies" (IETF, 2015) — Defines SFU, MCU, and other RTP topologies
2. RFC 8853 - "Using Simulcast in SDP and RTP Sessions" (IETF, 2021) — Simulcast negotiation that Go-Live's single-publisher model avoids
3. I-D.ietf-wish-whip - "WebRTC-HTTP Ingestion Protocol (WHIP)" — The ingestion protocol Go-Live implements
4. I-D.ietf-wish-whep - "WebRTC-HTTP Egress Protocol (WHEP)" — The egress protocol Go-Live implements
5. RFC 3550 - "RTP: A Transport Protocol for Real-Time Applications" (IETF, 2003) — The underlying transport protocol